Decision-centered wearable biosensors for personalized rehabilitation: integrating multimodal monitoring, artificial intelligence, and closed-loop intervention

Personalized rehabilitation requires repeated assessment of movement, physiological tolerance, fatigue, and adherence, yet conventional clinic-based measurements provide only intermittent snapshots. The convergence of wearable biosensors with artificial intelligence, edge-cloud computing, and connected data infrastructures creates an opportunity to convert continuous multimodal signals into timely rehabilitation decisions. This critical narrative review synthesizes physiological, biochemical, electrophysiological, biomechanical, and behavioral sensing within a monitoring–interpretation–feedback–adaptation–outcome framework. We first evaluate analytical validity, real-world robustness, biocompatibility, wearability, calibration drift, and power continuity. We then examine signal-quality control, multimodal fusion, digital biomarkers, federated learning, digital twins, and predictive models according to latency, uncertainty, generalizability, privacy, and decision utility. The review follows biosensor-derived information across a graded spectrum of action, from clinician-reviewed monitoring and real-time biofeedback to assist-as-needed adaptation, functional electrical stimulation, and state-responsive neuromodulation. Applications include motor and physiological assessment, fatigue-aware dose adjustment, musculoskeletal and neurological rehabilitation, posture correction, sensor-rich prosthetics, and Internet of Things–enabled home rehabilitation. Across these settings, greater autonomy increases the evidence and safety requirements: models should expose confidence, systems should degrade safely when signal quality falls, and digital infrastructures should preserve interoperability, version traceability, privacy, and accountable clinician oversight. The central translational gap is not sensor sensitivity alone but whether an integrated biosensor system produces a reliable metric, supports an explicit and proportionate decision, and improves an outcome important to the patient. Decision-centered evaluation can help move wearable biosensing from fragmented technology demonstrations toward secure, personalized, continuous, and clinically accountable rehabilitation.